Cloud Scaling

HS Tools in AWS

Application Load Balancer (smarter) routes to multiple EC2 instances looks inside request and assigns based on rules/roles path and host based routing Network LB (simplest, least expensive) routes based on low-level network distributes based on availability, not looking into request scales millions in seconds

ALB components

Load Balancer: entry point for request/action Listener: use rules to map requests to services in target groups Target Group: contains multiple EC2 instances

EC2 Bursting good idea when?

The MCOB SQL.business.nd.edu: Is burstable Sits idle most of the time, but then during class and before assignments it is much higher demand The credits of "use" and "compute" are collected overnight and then are used during peak hours, so that we buy a much smaller server than we would need for peak time

vertical scaling

add more resources to a machine to accomplish add'l work

horizontal scaling details

add resources at the same layer of a system, need a load balancer may need application reconfiguration may need software changes generally less expensive

scalability

adding/removing resources to a system to accommodate changes in demand

scaling lab

create a new instance in a different AZ create an application load balancer to distribute requests across the two servers create auto scaling group with minimum of 1 servers reasons not to use auto scaling: different regions not possible, really quick but not lasting for more than a few minutes, intricate launch config diff. b/w auto scaling and LB

horizontal scaling

distribute add'l work across additional machines

vertical scaling in AWS

identify bottleneck, identify new instance type, shut down server, select new type, start server

vertical constraints

memory, CPU, storage, network --> determine root cause to address in the scaling

reasons to scale vertically

monolithic applications legacy software no code change needed easy

downsides

need to turn off to scale designed for peak, so there are times you have too high, but need an outage to scale down

Autoscaling

scale programmatically identify scaling thresholds based on resource utilization monitor usage based on defined thresholds trigger scaling up/down when appropriate can have min # for config, so if failure, knows to start another back up right away

Bursting

vertical scaling on small scale w/o downtime bursts CPU only, not memory Limited to T3, T3a, and T2 types